simet.transforms.base.transform

simet.transforms.base.transform

Transform

Bases: ABC

Abstract interface for building torchvision transform pipelines.

Implementations should return a torchvision.transforms.Compose that defines the preprocessing/augmentation steps to apply to each sample (e.g., PIL Image → Tensor normalization).

Subclassing

Implement get_transform() to construct and return the composed transform. Keep any stochastic behavior (e.g., random crops) inside the returned pipeline, not in get_transform() itself.

Example

import torchvision.transforms as T class InceptionTransform(Transform): ... def get_transform(self) -> Compose: ... return T.Compose([ ... T.Resize(342), ... T.CenterCrop(299), ... T.ToTensor(), ... T.Normalize(mean=[0.485, 0.456, 0.406], ... std=[0.229, 0.224, 0.225]), ... ])

get_transform abstractmethod

get_transform()

Return the composed transform pipeline to apply per sample.

Returns:

Name	Type	Description
Compose	Compose	A torchvision Compose object encapsulating the
	Compose	preprocessing/augmentation steps.

Source code in simet/transforms/base/transform.py

@abstractmethod
def get_transform(self) -> Compose:
    """Return the composed transform pipeline to apply per sample.

    Returns:
        Compose: A torchvision `Compose` object encapsulating the
        preprocessing/augmentation steps.
    """
    pass